ZDFmediathek utilizes AI for personalized recommendations
ZDF is the second-largest public television channel in Germany, with a 15% market share (Statista 2024). To maintain its position as a leading online content provider, ZDF has focused on further developing the ZDFmediathek, ensuring the best possible offering for viewers.
Further development of the online offering
ZDF encountered various challenges in its further development. One aspect was the increased personalized distribution of content on its front ends, such as TV, apps, and third-party platforms like YouTube.
The development of automation systems and AI-supported recommendations, which increasingly expand the editorial content selection in the media library has become important.
In collaboration with ARD, ZDF is developing a joint streaming network that enables users to access and play ARD content in the ZDFmediathek. This requires close cooperation and technical integration of both system landscapes. The basis for all these efforts is the continual development of the complex system landscape and ZDF's data platform.
Algorithms in the ZDFmediathek: “This might interest you”
Utilizing algorithms, ZDFmediathek provides personalized and automated content recommendations based on viewer preferences.
The following figure displays the proportion of minutes watched by the use case “This might interest you” in the total viewing volume of the ZDFmediathek for automatically recommended content in the last 30 days.
We supported the ZDF team from researching algorithms to developing recommendation systems and integrating them into the existing system landscape, as well as operating them in the cloud environment. We also support the continuous evolution of the system architecture.
Algorithms are used in various areas:
- “This might interest you” suggests content that matches their interests in the ZDFmediathek
- “Also interesting” for recommending similar contributions to the content of the reference contribution
- “Stage” with recommendations on some stage positions
- “Because you watched ‘contribution’” suggests thematically similar recommendations to the video you watched
The international project team
“A mixed team of ZDF employees and currently 9 Accsonauts from South Africa and Germany are being deployed to solve the diverse tasks in this project. In this way, we combine a wide range of skills that complement each other perfectly and demonstrate our extensive expertise as a digital partner.”Dr. Volker JungPartner Accso
Accso covers the following tasks in the overall project:
- Architecture consulting and further development of the ZDFmediathek and neighboring systems
- Data engineering, architecture consulting, design, development and operation of the in-house data platform to enable easy exchange of data from multiple sources/departments/systems
- Data science and R&D (research and development) for recommendation algorithms
- Software development of the recommendation service and machine learning algorithms
- DevOps/MLOps and cloud operation in AWS and the Google Cloud
- Test management and quality assurance
